EECE 4353 Image Processing Vanderbilt University School of Engineering EECE/CS 4353 Image Processing
Lecture Notes: Histogram Point Processing of Images
Richard Alan Peters II
Department of Electrical Engineering and Computer Science
Fall Semester 2018
EC This work is licensed under the Creative Commons Attribution-Nonc mmercial 2.5L copy of this lic isit http://creativecommons.org/licenses/by-nc/2.5/ or end a letter to Cre ative Commons, 543 Howa PON a r, San m PUMA s USA.
EECE 4353 Image Processing Vanderbilt University School of Engineering Point Processes: Histogram Equalization
Task: remap a 1-band image I so that its histogram 1s as close to constant as possible. This maximizes the contrast evenly across the entire intensity range.
Let P(y+1) be the cumulative (probability) distribution function of I. Then J has, as closely as possible, a flat (constant) histogram 1f:
J(r,c,b) = 255- P| I(r,c,b)+1 |. ‘one-band The scaled CDF itself is used as the LUT. image That is, to equalize a one-band image, map it through its own CDF
multiplied by the maximum desired output value.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 2
EECE 4353 Image Processing Vanderbilt University School of Engineering Point Processes: Histogram Equalization
Task: remap image I with min = m, and max = M, so that its histogram is as close to constant as possible and has min = m; and max = M}.
Let P(y+1) be the cumulative (probability) distribution function of I.
Then J has, as closely as possible, the correct histogram 1f
Using ALAr.c) +1 |- Him +1 intensity | J(r.c) - (M, d extrema BE 2
+ m.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 3
EECE 4353 Image Processing Vanderbilt University School of Engineering
The CDF (cumulative distribution) Histogram EQ x 255 is the LUT for remapping.
Av mee -— See ee t qa E aaa WM
DRA! vu Masi: n 39 0s ^ Lr
A es oor S = ^t = "
Histogram EQ'd Value Image
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 4
EECE 4353 Image Processing Vanderbilt University School of Engineering
The CDF (cumulative distribution) Histogram EQ x 255 is the LUT for remapping.
QO SS ae (_.- —_— oe v '
Aa ah vu Masi: n 39 0s ^ Lr
A es oor S = ^t = "
Histogram EQ'd Value Image
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 5
EECE 4353 Image Processing Vanderbilt University School of Engineering
The CDF (cumulative distribution) Histogram EQ x 255 is the LUT for remapping.
ue Wes. oe Histogram EQ'd Value Image
EECE 4353 Image Processing Vanderbilt University School of Engineering
V
Histogram EQ
ae ` rl udi Ao E d — D^ A ien .-— mn n il
> TY 2 M
100 150
Luminance Histogram of Equalized Image
0 50 100 150
200
x =" _ TV aa ae m. Histogram EQ'd Value Image
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 7
Once again, but bigger EECE 4353 Image Processing (and with more feeling!)
Vanderbilt University School of Engineering
Histogram EQ of a Grayscale Image
x 10* Luminance Histogram of Kinkaku-ji
i
0 50 100 150 200 250
Luminance Image Luminance Histogram
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 8
NE EECE 4353 Image Processing (and with more feeling!) Vanderbilt University School of Engineering
Histogram EQ of a Grayscale Image
Luminance CDF of Kinkaku-ji
0 50 100 150 200 250
Luminance Image Luminance CDF
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 9
NE EECE 4353 Image Processing (and with more feeling!) Vanderbilt University School of Engineering
Histogram EQ of a Grayscale Image
LUT for Luminance Equalization of Kinkaku-ji 250
200 150
100
0 100 150 200 250
Luminance Image Equalization LU T
11 September 2018 (c) 1999-2018 by Richard Alan Peters II
Once again, but bigger EECE 4353 Image Processing (and with more feeling!)
Vanderbilt University School of Engineering
Histogram EQ of a Grayscale Image
x 10* Histogram of Equalized Kinkaku-ji Luminance image
0 0
Equalized Luminance Image Histogram of Eqd Image
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 11
Once again, but bigger EECE 4353 Image Processing (and with more feeling!)
Vanderbilt University School of Engineering
Histogram EQ of a Grayscale Image
Original Color Image Color Histograms
Here we will create a separate equalization LUT for each of the color bands from each of their CDFs.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 13
V One CDF-based EECE 4353 Image Processing LUT for each band Vanderbilt University School of Engineering
Histogram EQ of Individual Color Bands
CDFs of Kinkaku-ji Color Bands
Original Color Image Color CDFs
Here we will create a separate equalization LUT for each of the color bands from each of their CDFs.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 14
V One CDF-based EECE 4353 Image Processing LUT for each band Vanderbilt University School of Engineering
Histogram EQ of Individual Color Bands
LUT for Color Equalization of Kinkaku-ji
These are the separate equalization LUTs — one for each of the color bands.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 15
V One CDF-based EECE 4353 Image Processing LUT for each band Vanderbilt University School of Engineering
Histogram EQ of Individual Color Bands
x 10* Histogram of Color Equalized Kinkaku-ji
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 16
One CDF-based EECE 4353 Image Processing LUT for each band Vanderbilt University School of Engineering
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 17
One CDF-based EECE 4353 Image Processing LUT for each band Vanderbilt University School of Engineering
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 18
EECE 4353 Image Processing Vanderbilt University School of Engineering
Luminance EQ of a Color Image
Histograms of Kinkaku-ji Color Bands
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 19
EECE 4353 Image Processing Vanderbilt University School of Engineering
Luminance EQ of a Color Image
x 10° Luminance Histogram of Kinkaku-ji
0 50 100 150 200 250
Luminance Histogram
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 20
EECE 4353 Image Processing Vanderbilt University School of Engineering
Luminance EQ of a Color Image
CDFs of Kinkaku-ji Color Bands
Original Color Image Color CDFs
Here we compute the CDF of the luminance or value image, make a LUT from it, then remap all three color bands through the same LUT.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 21
EECE 4353 Image Processing Vanderbilt University School of Engineering
Luminance EQ of a Color Image
Luminance CDF of Kinkaku-ji
0 50 100 150 200 250
Luminance Image Luminance CDF
Here we compute the CDF of the luminance or value image, make a LUT from it, then remap all three color bands through the same LUT.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 22
EECE 4353 Image Processing Vanderbilt University School of Engineering
Luminance EQ of a Color Image
LUT for Luminance Equalization of Kinkaku-ji
0 100 150 200 250
Original Color Image Equalization LUT
Here we compute the CDF of the luminance or value image, make a LUT from it, then remap all three color bands through the same LUT.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 23
EECE 4353 Image Processing Vanderbilt University School of Engineering
Luminance EQ of a Color Image
This is what we get when we map each of the three bands through the LUT created from the CDF of the image’s luminance.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 24
EECE 4353 Image Processing Vanderbilt University School of Engineering
Histogram EQ of the Individual Bands
I ]
EL ee ee Oe BELEEELELICUIN
Each band EQ'd Separately Luminance EQ'd Color Image
The left image is what we got when we equalized each of the three bands separately. The right image 1s what we get when we map each color band through the luminance CDF
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 29
EECE 4353 Image Processing Vanderbilt University School of Engineering
Luminance EQ of a Color Image
s » m 6 l À v. = m» wr" » r , € E 7 " » = A é s = — a ya a o N Pol T." a F 4. = (i = 74 = aT
Original Color Image Each band EQ'd Separately
Remapping each of the color bands through the luminance CDF LUT preserves the hues of the original image better than the individual equalization of each color band.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 26
11 September 2018
EECE 4353 Image Processing Vanderbilt University School of Engineering
Point Processing: Histogram Matching
(c) 1999-2018 by Richard Alan Peters II 21
EECE 4353 Image Processing Vanderbilt University School of Engineering Point Processing: Histogram Matching
Task: Remap image I, (discolored) so that it has, as closely as possible, the same histogram as image J, (reference).
Q: Why do this?
A: Restore a degraded image based on an original. Match the characteristics of 1mages of the same scene from different cameras.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 28
EECE 4353 Image Processing Vanderbilt University School of Engineering Point Processing: Histogram Matching
Task: Remap image I, (discolored) so that it has, as closely as possible, the same histogram as image J, (reference).
Because the images are digital 1t 1s not, 1n general, possible to make h, = A.. Therefore, p, £ p, .
Q: How, then, can the matching be done? A: By matching percentiles.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 29
V d: discolored EECE 4353 Image Processing r: reference | Vanderbilt University School of Engineering
.. assuming a 1-band image,
Matching Percentiles one band of a color image
or its luminance image. Recall:
e CDF P, of image I, 1s such that 0 € P4( g4) S I.
e P,( g4t1) =c means that c 1s the fraction of pixels in I, that have a value less than or equal to g,.
* 100c 1s the percentile of pixels 1n I, that are less than or equal tO Zy.
To match percentiles, replace all occurrences of value g, in image I, with value g, from image J, whose percentile in J. most closely matches the percentile of g} in image I,.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 30
d: discolored EECE 4353 Image Processing
Hr reference Vanderbilt University School of Engineering
.. assuming a 1-band image,
Matching Percentiles one band of a color image
or its luminance image.
So, to create an image, K, from image I, such that K
has nearly the same CDF as image J, do the following:
Example: | Ae) = S P. (5) = 0.65 P. (9) = 0.56 P. (10) = 0.67 K(rc)- 10
If L(zc) = g; then let K(nc) = g, where g, is such that
Pa (Za) > P.(g,-1) AND Pa (Za) < P. (gy).
^a 10 ~ 1.0 | | s 08 — 08 iN a A "m 0.6 0.6 04 04 02 | 02 0.0 0.0 : 0123 4 |5 |6 7 8 9 101112131415 g; 012345678 9 101111213 14 15 gj
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 31
V d: discolored EECE 4353 Image Processing r: reference Vanderbilt University School of Engineering .. assuming a 1-band image,
Histogram Matching Algorithm one band of a color image Se
[R,C] =size (Ij) ; This directly matches K= zeros (R,C) ; image I, to image J..
P (g, 1) : CDF of J}.
min J, P (g, *1) « P (ga 1) u ; = max
— Tl; r r?
uu m = minl,
iii M,= max l.
K=K+ ZW E
end L | s.) Maybe faster to use a LUT. See slide 74.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 92
EECE 4353 Image Processing Vanderbilt University School of Engineering
V
Example: Histogram Matching
"Discolored" Image with 16 intensity 0.30 values.
Discolored Image pdf
Image 0.25 probability density function,
pdf.
0.10
0.05 | Wp 0.00
0123450678 Ais... g
IJ $1919q URTV preyory Aq 1912?) ooedgs uosuyor VSVN W leg POY :oSeui
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 33
EECE 4353 Image Processing Vanderbilt University School of Engineering
Example: Histogram Matching
Discolored Image CDF C 1.0 Image
=
E
O
Probability Distribution 0.8 m 0.6 0.4 0.2 0.0 |
01234567 8 9101112131415 S
II $1939q ue[y pueuorg Aq 193uo?) ooedg uosuyos VSVN W Med 19»[oos] :ogeui
*a.k.a Cumulative Distribution Function, CDF}.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 34
V EECE 4353 Image Processing Vanderbilt University School of Engineering
Example: Histogram Matching
Reference Reference Image pdf ima ge with 16 intensity te) 0.15 [^ Target values. so probability density function, C 0.12 + pdf;
Target a.k.a. Reference
0.09
0.06 |
0.03 | | 0.00 -
01234567289101112131415 g
/di[o-s1osnod-Su0JA /310A/u100 ueUIDpIe? MMM //:sdyy *c66] t'uoneuruv ueumpieV. ‘Weg YIN Å SIONO, FUOM ou, WOI [[Hs uoronpoujg :oSeuiq
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 35
V EECE 4353 Image Processing Vanderbilt University School of Engineering
Example: Histogram Matching
Reference Image CDF Rs is Target a Ha Probability Distribution Function, $ 0.8 - PDF,”
Target a.k.a. Reference
: 0.4 - i | 0.0 n £
01234567 89101112131415 g P Te
*a.k.a Cumulative Distribution Function, CDF).
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 36
EECE 4353 Image Processing Vanderbilt University School of Engineering Histogram Matching with a Lookup Table
The algorithm on slide 69 matches one image to another directly. Often it is faster or more versatile to use a lookup table (LUT). Rather than remapping each pixel in the image separately, one can create a table that indicates to which target value each input value should be mapped. Then
K = LUT[I, + 1]
In Matlab 1f the LUT 1s a 256 x | matrix with values from 0 to 255 and 1f image I, is one or multi-band of type uint8, it can
be remapped with the following code:
K = uint8(LUT(I-1));
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 37
EECE 4353 Image Processing Vanderbilt University School of Engineering Histogram Matching with a Lookup Table
The E-Z teenage New York version" on the previous page only works for one-band images. For truecolor or other multiband images you need to execute the LUT on each band separately. Viz:
Not true now! In : Nevertheless, I still Matlab 2018 e d d K : uintB(LU TE D) se % multiband LUT; multi-band image i A M works for both 1- fo = oe Page and watch Terry
and 3-band images. DUUM c IBgHSese eer tud pg Bozzio shred it. J(:,:,b) = LUT1D(1+double(I(:,:,b)));
end end
“http://globalia.net/donlope/fz/lyrics/Zappa_|In_ New_York.html#Page and https://www.youtube.com/watch?v-AQt2inyxNNg
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 38
V l EECE 4353 Image Processing LUT Creation Vanderbilt University School of Engineering
Discolored ap “T Reference CDF 0.8 0.8 0.6 0.4 0.4 0.2 0.2 0.0 0.0 z |
0 | 2 3(4)5 6 7 8 9 1011121314 15 012345067090 9 4011 12 13 14 15
15 LUT 12
012 3@)5 6 7 8 9 1011 12 13 14 15
Me
CO
UJ
e
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 39
Vanderbilt University School of Engineering
d: discolored EECE 4353 Image Processing r: reference
Look Up Table for Histogram Matching
LUT = zeros (256,10) ; This creates a look-up
g = 0; table which can then be used to remap the image. For g, = 0 to 255 l -
while P(g.+I<P,(g,+1) AND g «255 E TRSOEL | u P,(g,*1): CDF of L,
LUT( g, *1)- g;; P(g, +1): CDF of J,, LUT ( g,t+ 1) : Look-Up Table
end
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 40
Discolored & Reference CDFs, LUT, and Resultant CDF
11 September 2018
0
2
3
4
5
6
7 8 9 101112131415 2g
7 8
9
10 1112131415 g
P(g)
P(g)
EECE 4353 Image Processing Vanderbilt University School of Engineering
Ref. |
012 a o 7 8 9 101112131415 g
Result |
012345 6 7 8 Ail
(c) 1999-2018 by Richard Alan Peters II 41
EECE 4353 Image Processing Vanderbilt University School of Engineering
V
Example: Histogram Matching
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 42
EECE 4353 Image Processing Vanderbilt University School of Engineering
red pdf green pdf
blue pdf luminosity pdf
Image: Rocket Park at NASA Johnson Space Center by Richard Alan Peters II
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 43
EECE 4353 Image Processing Vanderbilt University School of Engineering
Cumulative Distribution Functions (CDF)
11 September 2018
Atlas-Mercury r, g, b, & v PDFs (Distributions)
50 100 150 200 250 1 Lf 0 50 100 150 200 250 1 I —— ———Á— i —ÀÀ 0 50 100 150 200 250 1 bo i 50 100 150 200 250 red CDF blue CDF green CDF luminosity CDF
Image: Rocket Park at NASA Johnson Space Center by Richard Alan Peters II
(c) 1999-2018 by Richard Alan Peters II 44
EECE 4353 Image Processing Vanderbilt University School of Engineering
V
red pdf green pdf
blue pdf luminosity pdf
Image: Production still from “The Wrong Trousers," Nick Park, Aardman Animation, 1993, https://www.aardman.com/work/wrong-trousers-clip/
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 45
V
EECE 4353 Image Processing Vanderbilt University School of Engineering
Cumulative Distribution Functions (CDF)
TechnoTrousers r, g, b, & v PDFs (Distributions)
50 100 150 200 250 1 eS a Jj BED 0 50 100 150 200 250 0 50 100 150 200 250 1 0 50 100 150 200 250 red CDF blue CDF green CDF luminosity CDF
Image: Production still from “The Wrong Trousers,” Nick Park, Aardman Animation, 1993, https://www.aardman.com/work/wrong-trousers-clip/
11 September 2018
(c) 1999-2018 by Richard Alan Peters II 46
EECE 4353 Image Processing Vanderbilt University School of Engineering
Remap one Image to have the RGB CDF of Another
2)
target "reference" R, G, & B remapped
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 47
EECE 4353 Image Processing Vanderbilt University School of Engineering
RGB CDks and the LUTSs
Atlas-Mercury Red PDF Atlas-Mercury Green PDF Atlas-Mercury Blue PDF
1 1 i 1
05r 4 05+ ,U GS 4 n 1 T
TechnoTrousers Red PDF 250 TechnoTrousers Green PDF 250 TechnoTrousers Blue PDF 250
1 1 1 0.5} + 05+ + 05+ - LUT (Red) Atlas-Mercury to TechnoTrousers 0 LUT (Green) Atlas-Mercury to TechnoTrousers LUT (Blue) Atlas-Mercury to TechnoTrousers 0
Atlas-Mercury RGB Remapped Red PDF 250 Atlas-Mercury RGB Remapped Green PDF 9 Atlas-Mercury RGB Remapped Blue PDF 250 | OO uem | 05+ 4 05. ee j 05+ J "C pee 0 (jene (J — — — — —L — — — ÉL. 50 100 150 200 250 50 100 150 200 250 50 100 150 200 250
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 48
VW EECE 4353 Image Processing Vanderbilt University School of Engineering
Effects of RGB Remapping on pdfs
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 49
WV EECE 4353 Image Processing Vanderbilt University School of Engineering
Effects of RGB Remapping on CDFs
efore After B
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 50
EECE 4353 Image Processing Vanderbilt University School of Engineering
Remap one Image to have the Lum. CDF of Another
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 5]
EECE 4353 Image Processing Vanderbilt University School of Engineering
Luminance CDFs and the LUT
Atlas-Mercury Luminosity CDF
0.5 0 90 TechnoTrousers Luminosity CDF 200 250 1 0.5 0 LUT (Luminosity) Atlas-Mercury to TechnoTrousers 250 200 100 0 50 Atlas-Mercury Remapped Luminosity CDF 200 250 1 E — 0.5 0 50 100 150 200 250
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 52
VW EECE 4353 Image Processing Vanderbilt University School of Engineering
Effects of Luminance Remapping on pdfs
100 150
“cl NEN . cutie, j
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 53
EECE 4353 Image Processing Vanderbilt University School of Engineering
Effects of Luminance Remapping on CDFs
Atlas-Mercury r, g, b, & v PDFs (Distributions)
Atlas-Mercury Value Remapped PDFs (Distributions)
50 100 150 200 250 50 100 150 200 250 1 1 0.5 0.5 NN ll 0 0 50 100 150 200 250 50 100 150 200 250 1 oS 1 — 0 0 50 100 150 200 250 50 100 150 200 250 1 ————— 1 SS ae . BEP 0 0 50 100 150 200 250 50 100 150 200 250 11 September 2018 (c) 1999-2018 by Richard Alan Peters II 54
EECE 4353 Image Processing Vanderbilt University School of Engineering
Histogram Equalization Revisited
Our first attempt to equalize the color version of Kinkaku-ji by mapping each band through its own CDF LUT led to the unsatisfactory results below (left):
|
55021 L n (a) Direct equalization of the color bands by (b) Equalization of the color bands by mapping mapping each band though its own CDF. each band though the CDF of the luminance LUT, = 255*CDF, image. LUT, = 255*CDF,.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 33
EECE 4353 Image Processing Vanderbilt University School of Engineering Histogram Equalization Revisited
Histogram matching presents another alternative: Match each color band CDF with the CDF from the luminance image. Then we get:
CIRS T =E
(a) Original image. (b) Equalization by mapping band, b, though LUT, = 255*[CDF,]'.*CDF,.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 56
Each color band CDF i tched
tote lonimence CbR te" | EECE 4353 Image Processing generate a LUT for each band Vanderbilt University School of Engineering
EQ by matching each color-band to image luminance
In 6 (count em) 1. Convert image I into grayscale image, L, via your 6 E-Z steps! favorite weighting scheme.
2. Compute the 3 color histograms, hc, C e {R,G,B} of I and the histogram, 7; , of L.
3. Compute the 4 probability density functions, pc, Ce {R,G,B,L}, from the he.
4. Compute the 4 cumulative distribution functions, He, Ce {R,G,B,L}, from the po.
9. Generate 3 lookup tables, Te, Ce {R,G,B}, by matching H, to H,, Ho to H,, & Hp to H}.
6. Map each image band {R,G,B! through its corresponding lookup table To.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 2d
Each color band CDF is matched . V to the luminance CDF to EECE 4353 Image Processing
generate a LUT for each band Vanderbilt University School of Engineering
EQ by matching each color-band to image luminance
Original Color Image Color & Lum. Histograms
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 58
generate a LUT for each band Vanderbilt University School of Engineering
Each color band CDF is matched . V to the luminance CDF to EECE 4353 Image Processing
EQ by matching each color-band to image luminance
CDFs of Kinkaku-ji Color Bands and Lumninance Image
100 150 200 250
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 59
generate a LUT for each band Vanderbilt University School of Engineering
Each color band CDF is matched . V to the luminance CDF to EECE 4353 Image Processing
EQ by matching each color-band to image luminance
Matched LUT for Luminance Remapping of Color Bands
0 100 150 200 250
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 60
Each color band CDF is matched . Y to the luminance CDF to EECE 4353 Image Processing
generate a LUT for each band Vanderbilt University School of Engineering
EQ by matching each color-band to image luminance
| TM
UN | Jj In 3c ze A ths
I | 2
| eil Luminance matched image Histo of lum matched image
Note the desaturation of the colors. |
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 61
Each band in one image is siii uo Gn nce aimee ded EECE 4353 Image Processing
in the other image.
Vanderbilt University School of Engineering
Right image: equalized each of the three bands separately (pp. 47-52).
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 62
P y Luminance is used to generate EECE 4353 Image Processing y one LUT for all bands. Vanderbilt University School of Engineering
Method 2: Map each band through 1 luminance LUT
o - a ^ - uet - s E " toT
A
a”
m.
Original color image Each band eqd through lum.
Right image: mapped each band through the LUT from luminance CDF (pp. 53-58).
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 63
Each color band CDF is matched VERAS EECE 4353 Image Processing generate a LUT for each band Vanderbilt University School of Engineering
Method 3: Match each band to luminance, 3 LUTs
Right image: matched each band to the luminance (pp. 90-95).
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 64
EECE 4353 Image Processing Vanderbilt University School of Engineering
V
Histogram Matching for Image Restoration
Ape. SG UPA AST 7
» IM TLCLE
, Nu. Fy A £v h
Lotus Flowers at Turtle Head Park, Lake Tai, Wuxi, Jiangsu Province, China. ib TE ELA AA Zia ILAA FE. Photos by R. A .Peters ll, July 2013.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 65
V EECE 4353 Image Processing Vanderbilt University School of Engineering
Histogram Matching for Image Restoration
Histogram of Degraded Scenic Picture 2 x 10* Histogram of Scenic Picture 1 x 10* Histogram of Remapped Degraded Image T T T T T T T
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 66
EECE 4353 Image Processing Vanderbilt University School of Engineering
V
Histogram Matching for Image Restoration
Lotus Flowers at Turtle Head Park, Lake Tai, Wuxi, Jiangsu Province, China. ib TE ELA Ail Zia ILAA FE. Photos by R. A .Peters ll, July 2013.
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 67
V EECE 4353 Image Processing
Vanderbilt University School of Engineering
Histogram Matching for Image Restoration
Another image of the same scene, not degraded | Jo EOS UP EN (URS LER 1.2 - Xm
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 68
Robot's view of boxes to move. EECE 4353 Image Processing http:/ /www.universalrobotics.com/ Vanderbilt University School of Engineering
Histogram Matching for Stereo Color Correction
Left Image . Right Image
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 69
V EECE 4353 Image Processing Vanderbilt University School of Engineering
Histogram Matching for Stereo Color Correction
x 10 g g g T : T 2.5 2 + 1.5 1 7 5 N Jl i} «me P nan 0 L ——— M —— ÀÁ— ce) 0 50 100 150 200 250
o
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 70
Robot's view of boxes to move. EECE 4353 Image Processing http:/ /www.universalrobotics.com/ Vanderbilt University School of Engineering
Histogram Matching for Stereo Color Correction
Left Image . Right Image
11 September 2018 (c) 1999-2018 by Richard Alan Peters II 71
EECE 4353 Image Processing
V Robot's view of boxes to move. : http://www.universalrobotics.com/ | Vanderbilt University School of Engineering
Histogram Matching for Stereo Color Correction
Right Image
Left Image
f; —
Right image histogram matched to left image.
12
11 September 2018 (c) 1999-2018 by Richard Alan Peters II